Gokul Rajaram says product managers now set the customer need and the why while engineers, designers and PMs build together in code, as long-running agents became reliable.
Gokul Rajaram expects the designer and PM to engineer ratio to move from about 1:3 or 1:10 toward 1:20, and says PMs must own evaluations of non-deterministic software.
Gokul Rajaram argues that judgment is the one durable skill in AI, and that Zendesk is among the most endangered companies because it prices per seat.
AI Collapses Product Roles and Rewards Judgment Over Code
AI is erasing the lines between product roles and between software moats. When agents make building cheap, what lasts is judgment, a deep problem in a valuable workflow, and assets like network effects, money flows and unique relationships. Per-seat tools and ad middlemen face the most pressure.
Takeaways
Product roles are collapsing into one building team. Since long-running agents became reliable, product managers set the customer need and guard the why while everyone builds in code. Designer and PM ratios to engineers move toward 1:20, and PMs own evaluations of non-deterministic software.
Judgment is the one future-proof skill. AI can produce lots of code that may be wrong or only sometimes valuable. The best product people edit features down, and every feature gets tested against the customer behaviour change it is meant to create.
Start AI products with a deep workflow problem. A Fortune 500 CIO told Gokul he would build agents himself with horizontal tools from Gemini and ChatGPT Enterprise. Startups therefore need a compelling problem in a high-value workflow, plus durability such as scarce assets, control points, hardware or network effects.
Per-seat legacy software is most exposed. Zendesk is among the most endangered because it prices per seat. A customer could buy 20 seats instead of 50 with 30 AI agents beside it. Companies holding timeless data, such as NetSuite, are more insulated than Slack.
Stickiness comes from money, networks and migration pain. Toast has payments flowing through it and DoorDash has network effects. Migration is a real barrier, since Square built scripts to move point-of-sale customers. Systems of record are also cutting API access, with Slack blocking Glean.
Ad middlemen get squeezed as agents replace apps. Winners own a coveted surface with identity and intent data, or deliver outcomes at a set cost as AppLovin does. Middlemen on big platforms get squeezed, and consumers using agents instead of apps is the bigger threat.
Great operators shaped his product instincts. At Google he saw Larry Page's focus on technology and scale and decade-long payoffs like Street View, TPUs and Waymo. Zuckerberg's custom audiences idea came from Zynga's whale problem, and Dorsey's design taste shaped Square's risk-based approach.
Management and hiring habits
Meeting 15 direct reports once a week takes 15 hours, which leaves a manager little time for their own work.
Weekly CEO emails include top-of-mind sections, and a weekly all-hands is useful even at a company of 15 to 20 people.
Tap a name to go there · tap a line to see why they are linked
On this episode 2
Show 1
Work-project interviews test real ability, such as Tony Xu's DoorDash task of trying to acquire a thousand customers.
Gokul acts as board buddy for the head of product or head of engineering on boards he sits on.
Palantir offers outcome-based engagement. It promises to solve a customer's most important problem in six months and is paid only if it succeeds.